sgl-project/sglang · error · ValueError

Video placeholder count does not match video_data: {len(vide

Error message

Video placeholder count does not match video_data: {len(video_data)} video(s) given

What it means

Raised by process_mm_data_async when, after substituting all supplied videos into their placeholders, a video placeholder token still remains in input_text. The count of video markers must equal len(video_data).

Source

Thrown at python/sglang/srt/multimodal/processors/dots_note_omni.py:454

                        audio_cap=audio_cap,
                        audio_sr=audio_sr,
                        k_mode=k_mode,
                        max_new_tokens=max_new_tokens,
                    ),
                )
                total_content_items += len(content)
                total_frames += sum(item.get("type") == "image_url" for item in content)
                total_audio_segments += sum(
                    item.get("type") == "audio_url" for item in content
                )
                input_text, media = self._render_video_content(
                    input_text, question, video_index, content
                )
                video_media.update(media)

            leftover = self.video_placeholder_regex.search(input_text)
            if leftover is not None:
                raise ValueError(
                    "Video placeholder count does not match video_data: "
                    f"{len(video_data)} video(s) given"
                )
            input_text, image_data, audio_data = self._merge_video_media(
                input_text, image_data, audio_data, video_media
            )
            preprocess_elapsed = time.perf_counter() - preprocess_started
            logger.info(
                "[dots_mm] rid=%s video_preprocess elapsed=%.3fs "
                "expanded_frames=%d expanded_audio_segments=%d content_items=%d "
                "after_preprocess images=%d audios=%d",
                request_obj.rid,
                preprocess_elapsed,
                total_frames,
                total_audio_segments,
                total_content_items,
                len(image_data),
                len(audio_data),

View on GitHub (pinned to 0132848349)

Solutions

1. Make the number of video placeholders in the prompt equal len(video_data)
2. If a video is optional, remove its placeholder when absent
3. Validate client-side: count video tokens in the prompt against your video list before sending

Example fix

// before
prompt = "<video> <video>"  # 2 placeholders
video_data = [v1]           # 1 video
// after
prompt = "<video>"          # 1 placeholder
video_data = [v1]

import re
VIDEO_RE = re.compile(r'<your-model-video-placeholder>')  # match server's video_placeholder_regex
count = len(VIDEO_RE.findall(prompt))
assert count == len(video_data or []), f'{count} placeholders vs {len(video_data or [])} videos'

try:
    await processor.process_mm_data_async(...)
except ValueError as e:
    if 'Video placeholder count' in str(e):
        rebuild prompt so placeholder count equals len(video_data); resubmit

When it happens

Trigger: The prompt/template contains more video placeholder tokens (matched by self.video_placeholder_regex) than entries in video_data — e.g. template renders 2 video tags but only 1 video was sent.

Common situations: Multi-video prompts built by hand or by a template that always emits N video slots; or a video item was dropped (None/failed download) before reaching the processor.

Related errors

• seq must be positive, got {seq}
• k_mode must not be empty
• Image placeholder count does not match image_data
• Audio placeholder count does not match audio_data
• Unsupported video input type: {type(video_file)}

AI-assisted analysis of sgl-project/sglang@0132848349 (2026-08-28). Data as JSON: /api/errors/29b6188266a02561. Report an issue: GitHub.